Now that nspawn can run unprivileged off directory trees owned by
the new "foreign" UID range let's make sure homed actually allows
files owned by that range in the home directories.
This is not enough to make nspawn just work in homed home dirs
unfortunately though. that's because homed applies an idmapping, and
nspawn would need to then to take that idmapped mount and apply another
one, and the kernel simply doesn't support stacked idmapped mounts.
There's work ongoing to address that in the kernel.
However, this is a first step, and should be enough to make things just
work should the kernel eventually support stacked idmapped mounts.
An object json is a set of key and value pair, hence has its number of
element is multiple of two, and we know at least two elements there.
Follow-up for b6a2df6307.
Fixes CID#1590724
HACKING.md should first and foremost tell someone how to hack on
systemd, installing packages from OBS isn't the most likely section
a new contributor will be interested in, so let's move it further
down.
Then, we can avoid that files outside of sysfs are written by udev ATTR key.
This also makes
- logs failure in udev_resolve_subsys_kernel(),
- failure in sd_device_get_syspath() critical, as that should not happen,
- cache the value to be write when running on test mode, to make it
shown by OPTIONS="dump" or obtained by ATTR match token.
This adds the same sections we already have in the homectl --help blurb
also to the man page.
While we are at it, let's also add a new section for Authentication
related switches.
* 4447d2974d Update changelog for 257.2-3 release
* 4b1c65b905 libudev1: add udeb back to shlibs
* 1974e3d06e systemd-boot: always check that the boot entry is set, even with Shim is already installed
* 9a5eea9823 systemd-boot: use boot entry argument instead of installing as grub.efi on ESP
* df6efeed46 libsystemd-dev/libudev-dev: depend on libcap-dev
* 5673b771e1 signing template: add override for executable-not-elf-or-script
* 3f109637c4 Update changelog for 257.2-2 release
* 42f4afa605 Drop udeb packages
* c04f7f2b16 signing template: always set urgency to 'high'
* 9bd8b5228b Set SBAT info for upstream build
* 257ba8563b udev: link to libsystemd-shared when building with noudeb profile
* 8ca2b26678 Link systemctl against libsystemd-shared
* 1a4a8af0c2 Install jq for pkg.systemd.upstream too since the template packages are now built
* 6fd0d2698d signing template: fix Lintian warnings and errors
* c79d10bbaa Build template packages for pkg.systemd.upstream profile, for OBS builds
* 485a867438 d/t/upstream: take into account autopkgtest pinning
* c1b6e565e3 Update README.source in the signing-template
* 17d1b92d9f d/t/control: remove 'flaky' from tests-in-lxd
* 2a36f6f5e1 Do not install sd-resolved and drop breaks-testbed from fast tests
* a3cb52f8d0 Enable UEFI on loong64
* ad7a943023 Enable libseccomp on loong64 and hppa
* 9d24f84ed5 Update changelog for 257.2-1 release
* f47619c9f4 Drop all patches, merged upstream
* d4aa6545a6 Install new files for upstream CI
* 5775daa46e d/rules: support building in OBS from git
There's finally quota on tmpfs, hence let's use it to make it harder for
users to DoS the system by consuming all disk space in /tmp/ and
/dev/shm/.
This enforces a default limit of 80% quota of the backing fs for these
two dirs for users, but this can be overriden in the user record, if
desired.
This also adds two other interesting features:
1. mount units gain GracefulOptions= which takes optional mount options
that are added only if supported by the kernel. (this is used to enable
usrquota on /tmp/, if available.)
2. The PAM logic in service management now supports reading passwords
from service credentials and via the askpw logic. This used for make
testing easy (so that we can run0 into a homed user which strictly
requires a password).
We dropped the variable in the packaging specs for Arch to keep the
integration points as minimal as possible so let's stop using it in
the build script as well.
TEST-74-AUX-UTILS covers many subtests, as it's a catch-all job, and a few
need a VM to run. The job is thus marked VM-only. But that means in settings
where we can't run VM tests (no KVM available), the entire thing is skipped,
losing tons of coverage that doesn't need skipping.
Move the VM-only subtests to TEST-87-AUX-UTILS-VM that is configured to only
run in VMs under both runners. This way we keep the existing tests as-is, and
we can add new VM-only tests without worrying. This is how the rest of the
tests are organized.
Follow-up for f4faac2073
So far nspawn supported unpriv containers only if backed by a DDI. This
adds dir-based unpriv containers too.
To make this work this introduces a new UID concept to systemd: the
"foreign UID range". This is a high UID range of size 64K. The idea is
that disk images that are "foreign" to the local system can use that,
and when a container or similar is invoked from it, a transiently
allocated dynamic UID range is mapped from that foreign UID range via id
mapped mounts.
This means the fully dynamic, transient UID ranges never hit the disk,
which should vastly simplify management, and does not require that uid
"subranges" are persistently delegated to any users.
The mountfsd daemon gained a new method call for acquiring an idmapped
mount fd for an mount tree owned by the foreign UID range. Access is
permitted to unpriv clients – as long as the referenced inode is located
within a dir owned by client's own uid range.
Enforce the quota on these two tmpfs at the same place where we mount
the per-user $XDG_RUNTIME_DIR. Conceptually these are very similar
concepts, and it makes sure to enforce the limits at the same place with
the same lifecycle.
Sometimes it's nice being able to store dev_t as pointer values in
hashmaps/tables, instead of having to allocate memory for them and using
devt_hash_ops. After all dev_t is weird on Linux/glibc: glibc defines it
as 64bit entity (which hence appears as something we cannot encode in a
pointer value for compat with 32bit archs) but it actually is 32bit in
the kernel apis. Hence we can safely cut off the upper 32bit, and still
retain compat with all archs.
But let's hide this in new macros, and validate this is all correct via
a test.
So far, we supported two modes:
1. when running unpriv we'd get the mounts from mountfsd, and the userns
from nsresourced
2. when running priv we'd do the mounts/userns ourselves
This untangles this a bit, so that we can also use mountfsd/nsresourced
when running privilged.
I think this is generally a bit nicer, and probably something we should
switch to entirely one day, as it reduces the variety of codepaths.
With this patch the default behaviour remains unchanged, but by
selecting the new "managed" option for --private-users= the codepaths
via mountfsd/nsresourced can be explicitly requested even when running
with privs.
This is mostly just reworks that we check for arg_userns_mode !=
USER_NAMESPACE_MANAGED rather than arg_privileged for a number of
codepaths, but requires more fixes, too. The devil is in the details.
This adds a new "foreign" value to --private-users-ownership= which is a
lot like "map", but maps from the host's foreign UID range rather than from the
host's 0.
(This has nothing much to do with making unprivileged directory-based
containers work, it's just very handy that we can run privileged
contains with such a mapping too, with an easy switch)
This simply calls into mountfsd to acquire the root mount and uses it as
root for the container.
Note that this also makes one more change: previously we ran containers
directory off their backing directory. Except when we didn't, and there
were a variety of exceptions: if we had no privs, if we ran off a disk
image, if the directory was the host's root dir, and some others.
This simplifies the logic a bit: we now simply always create a temporary
directory in /tmp/ and bind mount everything there, in all code paths.
This simplifies our code a bit. After all, in order to control
propagation we need to turn the root into a mount point anyway, hence we
might just do it at one place for all cases.
systemd-mountfsd so far provided a MountImage() API call for mounting a
disk image and returning a set of mount fds. This complements the API
with a new MountDirectory() API call, that operates on a directory
instead of an image file. Now, what makes this interesting is that it
applies an idmapping from the foreign UID range to the provided target
userns – and in which case unpriveleged operation is allowed (well,
under some conditions: in particular the client must own a parent dir of
the provided path).
This allows container managers to run fully unprivileged from
directories – as long as those directories are owned by the foreign UID
range. Basic operation is like this:
1. acquire a transient userns from systemd-nsresourced with 64K users
2. ask systemd-mountfsd for an idmapped mount of the container dir
matching that userns
3. join the userns and bind the mount fd as root.
Note that we have to drop various sandboxing knobs from the mountfsd
service file for this to work, since the kernel's security checks that
try to ensure than an obstructed /proc/ cannot be circumvented via
mounting a new procfs will otherwise prohibit mountfsd to duplicate the
mounts properly.
We currently set this at two distinct places right before calling
userdb_connect(). let's do this inside of userdb_connect() instead, and
derive it directly from the socket path.
This doesn't change behaviour but simplifies things a bit.
The GetUserRecord() and GetMemberships() have quite different arguments,
hence let's use separate structures for both.
This makes sense on its own, since it makes the structures a bit
smaller, but is also preparation for a later commit that adds a bunch of
new fields to one of the structs but not the other.
Lennart Poettering
Yu Watanabe
Daan De Meyer
Luca Boccassi
Mike Yuan